The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional markdown and configuration for academic research support. It contains no executable code or malicious logic.
[EXTERNAL_DOWNLOADS]: The skill references established academic databases and metadata services (EBSCO, Crossref, OpenAlex, ACM, IEEE). These are well-known, trusted resources for research and do not represent a security risk.
[SOURCE_INTEGRITY]: The skill includes a dedicated source-integrity.md file designed to prevent AI hallucinations by enforcing status tagging (User Provided, Verified, Unverified) and prohibiting the fabrication of citations.
[INDIRECT_PROMPT_INJECTION]: The skill processes external research data and user queries, which is an inherent attack surface. However, the risk is mitigated by the following:
Ingestion points: User research questions and external bibliographic data processed via SKILL.md and review-workflow.md.
Boundary markers: The agent is instructed by source-integrity.md to use explicit status labels to distinguish between processed inputs and verified knowledge.
Capability inventory: No high-risk tools (network exfiltration, file system write, or shell execution) are requested or utilized by the instructions.
Sanitization: Guidelines in source-integrity.md and output-contract.md explicitly prohibit fabricating data and require tagging unverified content in a dedicated 'to-verify' list.
[NO_CODE]: The skill does not bundle or execute any scripts, binaries, or shell commands. Instructions for maintenance scripts (scripts/sync_references.py) mentioned in discipline-routing.md are directed at the human developer for repository management, not the agent's runtime execution environment.

decompose-review-modules